The agent's last message is exactly the harness's final answer

spec:cc-agents.exact-answer

Use Case / Behavior · featureready floor ready

Packs: none

Design: 2 design entries · 2 pinned declarations · no decisions

specs/runner/exact-answer.sdp.md

1 open question and no blocking questions; implementation binding present, verifier binding present, and expected-outcome oracle none.

Intent

actor: A Workflow script or a parent agent that reads the report of a foreign agent.
problem: A Claude model that restates a foreign model's answer can alter it, and a Workflow reads the subagent's last message as the answer.
outcome: The turn ends with the harness's final text alone, byte for byte, and a failed turn says why.
value: A script can parse the answer, and a reader can trust it is the foreign model's own.

Open questions

non-blocking · question.endNotes line 23 · specs/runner/exact-answer.sdp.md:23 · An interrupted turn ends with a note that it was interrupted, and a completed turn that returned no text ends with a note saying so; both are built and no test pins their wording.Decider: delegatedProvisional: The notes are as built.

Behavior

rule.finalAlone line 29 · specs/runner/exact-answer.sdp.md:29: When the harness turn has ended, its final answer is delivered in a step of its own, as that step's only text, so a Workflow receives exactly it.
no example exercises it
rule.handbackWhereAsked line 30 · specs/runner/exact-answer.sdp.md:30: Where the agent's transcript tells it to hand its report back through SubagentHandback, the answer goes through that tool unchanged, and the turn counts as delivered only when the tool's result succeeded.
no example exercises it
rule.handbackTwiceRefused line 31 · specs/runner/exact-answer.sdp.md:31: A hand-back refused twice ends in the plain answer with the refusal appended, never a claim of delivery.
no example exercises it
rule.harnessTokens line 32 · specs/runner/exact-answer.sdp.md:32: No step reports usage to Claude Code; the harness's tokens appear as a text line naming the model, input, cached input and output tokens.
no example exercises it
rule.failureSaysWhy line 33 · specs/runner/exact-answer.sdp.md:33: A turn that did not complete ends with a bracketed note naming the harness, the status, the error kind and its message, plus a hint for quota, sign-in, capacity and rate-limit failures.
no example exercises it
rule.lateMessageGetsATurn line 34 · specs/runner/exact-answer.sdp.md:34: A message that reaches the agent as its harness turn ends is answered in a new turn before the hand-back.
no example exercises it

0 unkeyed rules.

Design

tokensLine line 38 · specs/runner/exact-answer.sdp.md:38: <model>: <n> input tokens (<n> cached), <n> output
no citation recorded
failureNote line 39 · specs/runner/exact-answer.sdp.md:39: [the <harness> turn <status> (<kind>): <message> <hint>]
no citation recorded

Verification

Verifier bindings

test:cc-agents.mod-suite · the mod against a scripted sidecar · anchored · test anchor binding present · plugin/tests/mod.test.ts:4

Design view

Requirement drawing

Declared verifying examples: no declared verifying example.

Requirement drawing for spec:cc-agents.exact-answer

The diagram as text

spec:cc-agents.exact-answer: The agent's last message is exactly the harness's final answer · declared Spec
spec:cc-agents.runner: Other model families run as real Claude Code agents · declared Spec
impl:cc-agents.mod: answers each step of an agent from the sidecar's events, hands back the exact answer, steers and stops, and returns the job state block · anchored code · plugin/hooks/register.js:9
test:cc-agents.mod-suite: the mod against a scripted sidecar · anchored test binding · binding present · plugin/tests/mod.test.ts:4
spec:cc-agents.exact-answer refines spec:cc-agents.runner · declared
impl:cc-agents.mod satisfies spec:cc-agents.exact-answer · anchored
test:cc-agents.mod-suite verifies spec:cc-agents.exact-answer · anchored

Mermaid source

Copyable source; the Studio draws static SVG.

Preferred and chosen syntax: requirementDiagram; exact refines, satisfies and verifies, with claim words in nodes and per-edge comments.

Mermaid source for Requirement drawing for spec:cc-agents.exact-answer

requirementDiagram
  requirement n0 {
    text: "The agent's last message is exactly the harness's final answer · declared Spec · spec:cc-agents.exact-answer"
  }
  element n1 {
    type: "answers each step of an agent from the sidecar's events, hands back the exact answer, steers and stops, and returns the job state block · anchored code · impl:cc-agents.mod"
  }
  element n2 {
    type: "the mod against a scripted sidecar · anchored test binding · binding present · test:cc-agents.mod-suite"
  }
  requirement n3 {
    text: "Other model families run as real Claude Code agents · declared Spec · spec:cc-agents.runner"
  }
  %% claim: anchored
  n1 - satisfies -> n0
  %% claim: declared
  n0 - refines -> n3
  %% claim: anchored
  n2 - verifies -> n0

Decisions

No decision record named.

Runtime

impl:cc-agents.mod · answers each step of an agent from the sidecar's events, hands back the exact answer, steers and stops, and returns the job state block · plugin/hooks/register.js:9
Uses
None.

Runtime observation: not tracked.

Evidence

Runtime observation: not tracked.

Build attestations: not tracked.

Deployments: not tracked.

Bills of materials: not tracked.

Impact

Relations of spec:cc-agents.exact-answer

The diagram as text

spec:cc-agents.exact-answer: The agent's last message is exactly the harness's final answer
spec:cc-agents.mod-layer: The mod answers an agent's steps from the sidecar and never calls a model
spec:cc-agents.runner: Other model families run as real Claude Code agents
spec:cc-agents.exact-answer dependsOn spec:cc-agents.mod-layer · declared — the final step is one of the steps the mod answers
spec:cc-agents.exact-answer refines spec:cc-agents.runner · declared

Mermaid source

Copyable source; the Studio draws static SVG.

Preferred syntax: flowchart. Chosen syntax: labelled flowchart, preserving exact relations, claims and folds.

Mermaid source for Relations of spec:cc-agents.exact-answer

flowchart TB
  n0["Other model families run as real Claude Code agents · spec:cc-agents.runner"]
  n1["The mod answers an agent's steps from the sidecar and never calls a model · spec:cc-agents.mod-layer"]
  n2["The agent's last message is exactly the harness's final answer · spec:cc-agents.exact-answer"]
  n2 -->|"dependsOn · declared — the final step is one of the steps the mod answers"| n1
  n2 -->|"refines · declared"| n0

Rests on

refines · spec:cc-agents.runner · stated defined, floor ready
dependsOn · spec:cc-agents.mod-layer · stated ready, floor ready

Rested on by

No Spec rests on this Spec.

Relations

Relation and direction	Other end	Claim	Reason	Exercised keys
dependsOn · outgoing	The mod answers an agent's steps from the sidecar and never calls a model `spec:cc-agents.mod-layer` · resolves	declared	the final step is one of the steps the mod answers	None
refines · outgoing	Other model families run as real Claude Code agents `spec:cc-agents.runner` · resolves	declared	None	None

Bindings

Implementation binding: present

impl:cc-agents.mod · answers each step of an agent from the sidecar's events, hands back the exact answer, steers and stops, and returns the job state block · plugin/hooks/register.js:9

Design reference: none

Verifier binding: present

test:cc-agents.mod-suite · the mod against a scripted sidecar · anchored · plugin/tests/mod.test.ts:4 · test anchor

Expected-outcome oracle: none

Runtime observation: not tracked.

Run results: not recorded in the graph.

Findings

None.

Composing (0)

Composing needs the script.